On 2015년 06월 10일 20:38, Marek Szyprowski wrote:
Hello,
On 2015-06-10 12:59, Inki Dae wrote:
Hi Marek,
On 2015년 06월 10일 19:03, Marek Szyprowski wrote:
Hello,
On 2015-06-01 17:04, Gustavo Padovan wrote:
From: Gustavo Padovan gustavo.padovan@collabora.co.uk
Hi,
Here goes the full support for atomic modesetting on exynos. I've split the patches in the various phases of atomic support.
Thanks for this patchses, however I've noticed a problem after applying them. The issue gets revealed when support for IOMMU is enabled. I've did my tests with Exynos HDMI driver on Odroid U3 board.
To demonstrated the issue I've added following additional debug in the exynos mixer driver in mixer_graph_buffer() function: pr_info("dma addr %pad plane->src_width %d plane->src_height %d\n", &plane->dma_addr[0], plane->src_width, plane->src_height);
Before applying patches setting 640x480 mode and getting back to 1920x1080 console generates following log:
# modetest -M exynos -s 23:640x480 setting mode 640x480-60Hz@XR24 on connectors 23, crtc 21 [ 3860.617151] dma 0xbc500000 plane->src_width 640 plane->src_height 480 ^C [ 3870.555232] dma 0xbbd00000 plane->src_width 1920 plane->src_height 1080 [ 3870.565696] dma 0xbbd00000 plane->src_width 1920 plane->src_height 1080
After applying atomic modesetting patchset: # modetest -M exynos -s 24:640x480 [ 142.540122] dma 0xbbd00000 plane->src_width 1920 plane->src_height 1080 [ 142.550726] dma 0xbbd00000 plane->src_width 1920 plane->src_height 1080 setting mode 640x480-60Hz@XR24 on connectors 24, crtc 22 [ 142.643672] dma 0xbc500000 plane->src_width 1920 plane->src_height 1080 [ 142.759982] dma 0xbc500000 plane->src_width 640 plane->src_height 480 ^C [ 154.986040] dma 0xbbd00000 plane->src_width 1920 plane->src_height 1080
As you can see from the above log, mixer_graph_buffer function is called several times. 0xbbd00000 is the DMA address of the 1920x1080 framebuffer and 0xbc500000 is the DMA address of the allocated 640x480 buffer. mixer_graph_buffer() is first called with the new DMA address of the framebuffer, but with the old mode parameters (1920x1080 size) and then in the next call it updates the plane parameters to the correct values (size changed to 640x480). When IOMMU is not used, this can be easily missed, but after enabling IOMMU support, any DMA access to unallocated address causes IOMMU PAGE FAULT. Here it will happen after changing DMA address of the buffer without changing the size.
A quick workaround to resolve this multiple calls to mixer_graph_buffer() with partially updated mode values is to remove calls to mixer_window_suspend/mixer_window_resume from mixer_disable and mixer_disable functions, but I expect that this is not the right approach.
Probably the same problem can be observed with Exynos FIMD driver.
Gustavo: could you check if mixer_enable functions should really call mixer_window_resume function, which in turn calls mixer_win_commit, which calls mixer_graph_buffer with partially updated display buffer data?
Marek, can you share how other people can test the atomic feature with iommu?
I should have merged below several patches and added device tree relevant codes to test iommu.
- Merged iommu support patches for Exynos SoC below iommu exynos tree,
https://git.kernel.org/cgit/linux/kernel/git/joro/iommu.git/log/?h=arm/exyno...
- Merged below patch, [PATCH v7 24/25] ARM: DMA-mapping: add
support for creating reserved mappings in iova space
- Added device node relevant codes - I tested Exynos drm on trats2
board based on Exynos4412 SoC - like below, in exynos4.dtsi file: fimd: fimd@11c00000 { ... iommus = <&sysmmu_fimd0>; ...
sysmmu_fimd0: sysmmu@11E20000 { compatible = "samsung,exynos-sysmmu"; reg = <0x11E20000 0x1000>; interrupt-parent = <&combiner>; interrupts = <5 2>; clock-names = "sysmmu", "master"; clocks = <&clock CLK_SMMU_FIMD0>, <&clock CLK_FIMD0>; power-domains = <&pd_lcd0>; #iommu-cells = <0>; };
in exynos4412-trats2.dts file: fimd@11c00000 { status = "okay"; iommu-reserved-mapping = <0x40000000 0x40000000 0x40000000>; };
Is that all? You would need to share exact guide about iommu enabling to other people so that they can test atomic feature with iommu.
Right, the above should be enough. For convenience I've prepared a branch with all needed patches: https://git.linaro.org/people/marek.szyprowski/linux-srpol.git v4.1-exynos-drm-iommu
I've included following branches: https://git.kernel.org/cgit/linux/kernel/git/daeinki/drm-exynos.git/log/?h=e...
https://git.kernel.org/cgit/linux/kernel/git/kgene/linux-samsung.git/log/?h=...
https://git.kernel.org/cgit/linux/kernel/git/kgene/linux-samsung.git/log/?h=...
https://git.kernel.org/cgit/linux/kernel/git/joro/iommu.git/log/?h=arm/exyno...
Then I've applied my Exynos DRM patches (http://www.spinics.net/lists/linux-samsung-soc/msg45114.html) as well as rebased patch 24/25 and 25/25 from my initial v7 IOMMU series (http://www.spinics.net/lists/linux-samsung-soc/msg44652.html).
Thanks, Marek.
Gustavo and Joonyoung, let's resolve the page fault issue with iommu this time. I'd like to merge most patch sets - atomic feature, iommu, Exynos5433 IP support, and driver initialization consolidating. I should have pull-request in the near feature. So we need to do this in a hurry.
Thanks, Inki Dae
Best regards